FLUX Models Comparison: From FLUX.1 to FLUX.2
Complete guide to FLUX AI image generation models - comparing FLUX.1 Pro, Dev, Schnell and the new FLUX.2 with multi-reference support and 32B parameters.
In the rapidly evolving landscape of AI-generated imagery, FLUX by Black Forest Labs has established itself as a leading force. With the recent release of FLUX.2 in November 2025, the lineup now spans two generations of models — each with distinct strengths for different use cases.
This guide covers both FLUX.1 and the new FLUX.2, helping you choose the right model for your needs.
What's New in FLUX.2
Released in November 2025, FLUX.2 represents a major leap forward with a 32 billion parameter architecture that unifies image generation and editing in a single model.
Key FLUX.2 Features
- Multi-Reference Support — combine up to 10 images into a novel output
- 4 Megapixel Editing — edit high-resolution images while preserving detail
- Superior Typography — handles complex text, infographics, and UI mockups with legible fine text
- Unified Architecture — generation and editing in one checkpoint
- Vision-Language Integration — couples with Mistral-3 24B VLM for enhanced understanding
FLUX.2 Model Variants
| Model | Description | Availability |
|---|---|---|
| FLUX.2 [pro] | Highest quality, matches leading closed-source systems | API, Playground |
| FLUX.2 [flex] | Adjustable parameters for speed/quality tradeoff | API, Playground |
| FLUX.2 [dev] | 32B open weights, unified generation + editing | Hugging Face, GitHub |
| FLUX.2 [klein] | Distilled model (coming soon) | Apache 2.0 |
FLUX.2 Benchmark Results
In head-to-head comparisons, FLUX.2 [Dev] leads all open-weight alternatives:
- 66.6% win rate in text-to-image generation
- 59.8% win rate in single-reference editing
- Outperforms Qwen-Image (51.3%) and Hunyuan Image 3.0 (48.1%)
Hardware Requirements
FLUX.2's 32B parameters require significant resources:
- Full precision: 90GB VRAM
- FP8 quantized: ~54GB VRAM (40% reduction, comparable quality)
FLUX.1 Models (Previous Generation)
While FLUX.2 is the latest, FLUX.1 models remain excellent choices — especially for users with limited hardware or specific licensing needs.
FLUX.1 Pro
The FLUX.1 Pro flagship model offers:
- Superior prompt-following capabilities
- Exceptional visual quality
- Excellent text rendering in images
- Diverse output styles
FLUX.1 Dev
The FLUX.1 Dev model is tailored for research and development:
- Open weights for inspection and modification
- Non-commercial license
- High quality output comparable to Pro
- Decent text rendering
FLUX.1 Schnell
For speed and flexibility, FLUX.1 Schnell ("fast" in German) delivers:
- Fastest generation in the FLUX.1 lineup
- Apache 2.0 license — fully open for any use
- Lower resource requirements
- Optimized for local/consumer hardware
Complete Comparison Table
| Feature | FLUX.2 Pro | FLUX.2 Dev | FLUX.1 Pro | FLUX.1 Dev | FLUX.1 Schnell |
|---|---|---|---|---|---|
| Parameters | 32B | 32B | 12B | 12B | 12B |
| Image Quality | Excellent | Excellent | Excellent | Very Good | Good |
| Multi-Reference | Yes (10 images) | Yes (10 images) | No | No | No |
| Image Editing | Yes | Yes | No | No | No |
| Max Resolution | 4MP | 4MP | 2MP | 2MP | 2MP |
| Text in Images | Excellent | Excellent | Excellent | Good | Limited |
| Speed | Medium | Medium | Slower | Medium | Fastest |
| License | Commercial API | Open weights | Commercial API | Non-commercial | Apache 2.0 |
| VRAM Required | API | 90GB / 54GB FP8 | API | ~24GB | ~12GB |
| Cost | Paid | Free | Paid | Free | Free |
Which Model Should You Choose?
Choose FLUX.2 if:
- You need multi-reference image combination
- Image editing capabilities are important
- You require the highest quality available
- You have access to high-end hardware (for Dev) or API budget (for Pro)
- Complex typography and infographics are part of your workflow
Choose FLUX.1 Pro if:
- You need commercial-ready output via API
- FLUX.2 API costs exceed your budget
- Your workflow doesn't require image editing features
Choose FLUX.1 Dev if:
- You're doing research or academic work
- You want to inspect/modify weights for non-commercial use
- You have moderate GPU resources (~24GB VRAM)
Choose FLUX.1 Schnell if:
- Speed is your top priority
- You need Apache 2.0 licensing freedom
- You're running on consumer hardware
- You're doing rapid prototyping
Conclusion
Black Forest Labs continues to push the boundaries of open AI image generation. FLUX.2 brings groundbreaking multi-reference support and unified generation/editing, while FLUX.1 models remain practical choices for users with different hardware constraints and licensing needs.
For most new projects with adequate resources, FLUX.2 is the clear choice. For quick local experimentation or Apache 2.0 licensing requirements, FLUX.1 Schnell remains unbeatable.
Sources: Black Forest Labs, NVIDIA Blog, Hugging Face